Further validation anomaly detection

Hey everyone,

Lisa and I created a short-term plan for the upcoming two weeks to further validate the anomaly detection. The points are prioritized as follows:

• Integrate shuffling in dataset (shuffle windows, but each window is still in order)
• integrate shuffling in every epoch

Test different architectures:

VAE:

• train with different parameters
• train with high window size and epochs
• train with ee pose and forces
• train with normalized data
• train with ee vel and forces

Conv AE:

• train with different parameters
• train with high window size and epochs
• train with ee pose and forces
• train with normalized data
• train with ee vel and forces

VAE-CNN:

• test new architecture
• train random model and validate

Record additional data:

• severe anomalies (free space movement + collision / different traj) regarding time and amplitude,
• more variance

Additional points to check:

• Train a prediction model
• check out "new" experimental data from VW clipping task, data available here and already classified with iO / niO.
• Integrate normalization
• Evaluate other loss_fn